2 * Copyright (C) 2006-2007 Red Hat, Inc. All rights reserved.
4 * This copyrighted material is made available to anyone wishing to use,
5 * modify, copy, or redistribute it subject to the terms and conditions
6 * of the GNU General Public License v.2.
9 #include <linux/miscdevice.h>
10 #include <linux/init.h>
11 #include <linux/wait.h>
12 #include <linux/module.h>
13 #include <linux/file.h>
15 #include <linux/poll.h>
16 #include <linux/signal.h>
17 #include <linux/spinlock.h>
18 #include <linux/dlm.h>
19 #include <linux/dlm_device.h>
21 #include "dlm_internal.h"
22 #include "lockspace.h"
24 #include "lvb_table.h"
27 static const char *name_prefix="dlm";
28 static struct miscdevice ctl_device;
29 static const struct file_operations device_fops;
33 struct dlm_lock_params32 {
47 char lvb[DLM_USER_LVB_LEN];
51 struct dlm_write_request32 {
58 struct dlm_lock_params32 lock;
59 struct dlm_lspace_params lspace;
60 struct dlm_purge_params purge;
71 struct dlm_lock_result32 {
77 struct dlm_lksb32 lksb;
80 /* Offsets may be zero if no data is present */
84 static void compat_input(struct dlm_write_request *kb,
85 struct dlm_write_request32 *kb32,
88 kb->version[0] = kb32->version[0];
89 kb->version[1] = kb32->version[1];
90 kb->version[2] = kb32->version[2];
93 kb->is64bit = kb32->is64bit;
94 if (kb->cmd == DLM_USER_CREATE_LOCKSPACE ||
95 kb->cmd == DLM_USER_REMOVE_LOCKSPACE) {
96 kb->i.lspace.flags = kb32->i.lspace.flags;
97 kb->i.lspace.minor = kb32->i.lspace.minor;
98 strcpy(kb->i.lspace.name, kb32->i.lspace.name);
99 } else if (kb->cmd == DLM_USER_PURGE) {
100 kb->i.purge.nodeid = kb32->i.purge.nodeid;
101 kb->i.purge.pid = kb32->i.purge.pid;
103 kb->i.lock.mode = kb32->i.lock.mode;
104 kb->i.lock.namelen = kb32->i.lock.namelen;
105 kb->i.lock.flags = kb32->i.lock.flags;
106 kb->i.lock.lkid = kb32->i.lock.lkid;
107 kb->i.lock.parent = kb32->i.lock.parent;
108 kb->i.lock.xid = kb32->i.lock.xid;
109 kb->i.lock.timeout = kb32->i.lock.timeout;
110 kb->i.lock.castparam = (void *)(long)kb32->i.lock.castparam;
111 kb->i.lock.castaddr = (void *)(long)kb32->i.lock.castaddr;
112 kb->i.lock.bastparam = (void *)(long)kb32->i.lock.bastparam;
113 kb->i.lock.bastaddr = (void *)(long)kb32->i.lock.bastaddr;
114 kb->i.lock.lksb = (void *)(long)kb32->i.lock.lksb;
115 memcpy(kb->i.lock.lvb, kb32->i.lock.lvb, DLM_USER_LVB_LEN);
116 if (kb->i.lock.namelen <= max_namelen)
117 memcpy(kb->i.lock.name, kb32->i.lock.name,
120 kb->i.lock.namelen = max_namelen;
124 static void compat_output(struct dlm_lock_result *res,
125 struct dlm_lock_result32 *res32)
127 res32->version[0] = res->version[0];
128 res32->version[1] = res->version[1];
129 res32->version[2] = res->version[2];
131 res32->user_astaddr = (__u32)(long)res->user_astaddr;
132 res32->user_astparam = (__u32)(long)res->user_astparam;
133 res32->user_lksb = (__u32)(long)res->user_lksb;
134 res32->bast_mode = res->bast_mode;
136 res32->lvb_offset = res->lvb_offset;
137 res32->length = res->length;
139 res32->lksb.sb_status = res->lksb.sb_status;
140 res32->lksb.sb_flags = res->lksb.sb_flags;
141 res32->lksb.sb_lkid = res->lksb.sb_lkid;
142 res32->lksb.sb_lvbptr = (__u32)(long)res->lksb.sb_lvbptr;
146 /* Figure out if this lock is at the end of its life and no longer
147 available for the application to use. The lkb still exists until
148 the final ast is read. A lock becomes EOL in three situations:
149 1. a noqueue request fails with EAGAIN
150 2. an unlock completes with EUNLOCK
151 3. a cancel of a waiting request completes with ECANCEL/EDEADLK
152 An EOL lock needs to be removed from the process's list of locks.
153 And we can't allow any new operation on an EOL lock. This is
154 not related to the lifetime of the lkb struct which is managed
155 entirely by refcount. */
157 static int lkb_is_endoflife(struct dlm_lkb *lkb, int sb_status, int type)
165 if (lkb->lkb_grmode == DLM_LOCK_IV)
169 if (type == AST_COMP && lkb->lkb_grmode == DLM_LOCK_IV)
176 /* we could possibly check if the cancel of an orphan has resulted in the lkb
177 being removed and then remove that lkb from the orphans list and free it */
179 void dlm_user_add_ast(struct dlm_lkb *lkb, int type)
182 struct dlm_user_args *ua;
183 struct dlm_user_proc *proc;
184 int eol = 0, ast_type;
186 if (lkb->lkb_flags & (DLM_IFL_ORPHAN | DLM_IFL_DEAD))
189 ls = lkb->lkb_resource->res_ls;
190 mutex_lock(&ls->ls_clear_proc_locks);
192 /* If ORPHAN/DEAD flag is set, it means the process is dead so an ast
193 can't be delivered. For ORPHAN's, dlm_clear_proc_locks() freed
194 lkb->ua so we can't try to use it. This second check is necessary
195 for cases where a completion ast is received for an operation that
196 began before clear_proc_locks did its cancel/unlock. */
198 if (lkb->lkb_flags & (DLM_IFL_ORPHAN | DLM_IFL_DEAD))
201 DLM_ASSERT(lkb->lkb_astparam, dlm_print_lkb(lkb););
202 ua = (struct dlm_user_args *)lkb->lkb_astparam;
205 if (type == AST_BAST && ua->bastaddr == NULL)
208 spin_lock(&proc->asts_spin);
210 ast_type = lkb->lkb_ast_type;
211 lkb->lkb_ast_type |= type;
214 kref_get(&lkb->lkb_ref);
215 list_add_tail(&lkb->lkb_astqueue, &proc->asts);
216 wake_up_interruptible(&proc->wait);
218 if (type == AST_COMP && (ast_type & AST_COMP))
219 log_debug(ls, "ast overlap %x status %x %x",
220 lkb->lkb_id, ua->lksb.sb_status, lkb->lkb_flags);
222 eol = lkb_is_endoflife(lkb, ua->lksb.sb_status, type);
224 lkb->lkb_ast_type &= ~AST_BAST;
225 lkb->lkb_flags |= DLM_IFL_ENDOFLIFE;
228 /* We want to copy the lvb to userspace when the completion
229 ast is read if the status is 0, the lock has an lvb and
230 lvb_ops says we should. We could probably have set_lvb_lock()
231 set update_user_lvb instead and not need old_mode */
233 if ((lkb->lkb_ast_type & AST_COMP) &&
234 (lkb->lkb_lksb->sb_status == 0) &&
235 lkb->lkb_lksb->sb_lvbptr &&
236 dlm_lvb_operations[ua->old_mode + 1][lkb->lkb_grmode + 1])
237 ua->update_user_lvb = 1;
239 ua->update_user_lvb = 0;
241 spin_unlock(&proc->asts_spin);
244 spin_lock(&proc->locks_spin);
245 if (!list_empty(&lkb->lkb_ownqueue)) {
246 list_del_init(&lkb->lkb_ownqueue);
249 spin_unlock(&proc->locks_spin);
252 mutex_unlock(&ls->ls_clear_proc_locks);
255 static int device_user_lock(struct dlm_user_proc *proc,
256 struct dlm_lock_params *params)
259 struct dlm_user_args *ua;
262 ls = dlm_find_lockspace_local(proc->lockspace);
266 if (!params->castaddr || !params->lksb) {
271 ua = kzalloc(sizeof(struct dlm_user_args), GFP_KERNEL);
275 ua->user_lksb = params->lksb;
276 ua->castparam = params->castparam;
277 ua->castaddr = params->castaddr;
278 ua->bastparam = params->bastparam;
279 ua->bastaddr = params->bastaddr;
280 ua->xid = params->xid;
282 if (params->flags & DLM_LKF_CONVERT)
283 error = dlm_user_convert(ls, ua,
284 params->mode, params->flags,
285 params->lkid, params->lvb,
286 (unsigned long) params->timeout);
288 error = dlm_user_request(ls, ua,
289 params->mode, params->flags,
290 params->name, params->namelen,
291 (unsigned long) params->timeout);
293 error = ua->lksb.sb_lkid;
296 dlm_put_lockspace(ls);
300 static int device_user_unlock(struct dlm_user_proc *proc,
301 struct dlm_lock_params *params)
304 struct dlm_user_args *ua;
307 ls = dlm_find_lockspace_local(proc->lockspace);
311 ua = kzalloc(sizeof(struct dlm_user_args), GFP_KERNEL);
315 ua->user_lksb = params->lksb;
316 ua->castparam = params->castparam;
317 ua->castaddr = params->castaddr;
319 if (params->flags & DLM_LKF_CANCEL)
320 error = dlm_user_cancel(ls, ua, params->flags, params->lkid);
322 error = dlm_user_unlock(ls, ua, params->flags, params->lkid,
325 dlm_put_lockspace(ls);
329 static int device_user_deadlock(struct dlm_user_proc *proc,
330 struct dlm_lock_params *params)
335 ls = dlm_find_lockspace_local(proc->lockspace);
339 error = dlm_user_deadlock(ls, params->flags, params->lkid);
341 dlm_put_lockspace(ls);
345 static int create_misc_device(struct dlm_ls *ls, char *name)
350 len = strlen(name) + strlen(name_prefix) + 2;
351 ls->ls_device.name = kzalloc(len, GFP_KERNEL);
352 if (!ls->ls_device.name)
355 snprintf((char *)ls->ls_device.name, len, "%s_%s", name_prefix,
357 ls->ls_device.fops = &device_fops;
358 ls->ls_device.minor = MISC_DYNAMIC_MINOR;
360 error = misc_register(&ls->ls_device);
362 kfree(ls->ls_device.name);
368 static int device_user_purge(struct dlm_user_proc *proc,
369 struct dlm_purge_params *params)
374 ls = dlm_find_lockspace_local(proc->lockspace);
378 error = dlm_user_purge(ls, proc, params->nodeid, params->pid);
380 dlm_put_lockspace(ls);
384 static int device_create_lockspace(struct dlm_lspace_params *params)
386 dlm_lockspace_t *lockspace;
390 if (!capable(CAP_SYS_ADMIN))
393 error = dlm_new_lockspace(params->name, strlen(params->name),
394 &lockspace, params->flags, DLM_USER_LVB_LEN);
398 ls = dlm_find_lockspace_local(lockspace);
402 error = create_misc_device(ls, params->name);
403 dlm_put_lockspace(ls);
406 dlm_release_lockspace(lockspace, 0);
408 error = ls->ls_device.minor;
413 static int device_remove_lockspace(struct dlm_lspace_params *params)
415 dlm_lockspace_t *lockspace;
417 int error, force = 0;
419 if (!capable(CAP_SYS_ADMIN))
422 ls = dlm_find_lockspace_device(params->minor);
426 /* Deregister the misc device first, so we don't have
427 * a device that's not attached to a lockspace. If
428 * dlm_release_lockspace fails then we can recreate it
430 error = misc_deregister(&ls->ls_device);
432 dlm_put_lockspace(ls);
435 kfree(ls->ls_device.name);
437 if (params->flags & DLM_USER_LSFLG_FORCEFREE)
440 lockspace = ls->ls_local_handle;
442 /* dlm_release_lockspace waits for references to go to zero,
443 so all processes will need to close their device for the ls
444 before the release will procede */
446 dlm_put_lockspace(ls);
447 error = dlm_release_lockspace(lockspace, force);
449 create_misc_device(ls, ls->ls_name);
454 /* Check the user's version matches ours */
455 static int check_version(struct dlm_write_request *req)
457 if (req->version[0] != DLM_DEVICE_VERSION_MAJOR ||
458 (req->version[0] == DLM_DEVICE_VERSION_MAJOR &&
459 req->version[1] > DLM_DEVICE_VERSION_MINOR)) {
461 printk(KERN_DEBUG "dlm: process %s (%d) version mismatch "
462 "user (%d.%d.%d) kernel (%d.%d.%d)\n",
464 task_pid_nr(current),
468 DLM_DEVICE_VERSION_MAJOR,
469 DLM_DEVICE_VERSION_MINOR,
470 DLM_DEVICE_VERSION_PATCH);
480 * dlm_user_request -> request_lock
481 * dlm_user_convert -> convert_lock
484 * dlm_user_unlock -> unlock_lock
485 * dlm_user_cancel -> cancel_lock
487 * device_create_lockspace
490 * device_remove_lockspace
491 * dlm_release_lockspace
494 /* a write to a lockspace device is a lock or unlock request, a write
495 to the control device is to create/remove a lockspace */
497 static ssize_t device_write(struct file *file, const char __user *buf,
498 size_t count, loff_t *ppos)
500 struct dlm_user_proc *proc = file->private_data;
501 struct dlm_write_request *kbuf;
502 sigset_t tmpsig, allsigs;
506 if (count < sizeof(struct dlm_write_request32))
508 if (count < sizeof(struct dlm_write_request))
512 kbuf = kmalloc(count, GFP_KERNEL);
516 if (copy_from_user(kbuf, buf, count)) {
521 if (check_version(kbuf)) {
527 if (!kbuf->is64bit) {
528 struct dlm_write_request32 *k32buf;
529 k32buf = (struct dlm_write_request32 *)kbuf;
530 kbuf = kmalloc(count + (sizeof(struct dlm_write_request) -
531 sizeof(struct dlm_write_request32)), GFP_KERNEL);
536 set_bit(DLM_PROC_FLAGS_COMPAT, &proc->flags);
537 compat_input(kbuf, k32buf,
538 count - sizeof(struct dlm_write_request32));
543 /* do we really need this? can a write happen after a close? */
544 if ((kbuf->cmd == DLM_USER_LOCK || kbuf->cmd == DLM_USER_UNLOCK) &&
545 test_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags))
548 sigfillset(&allsigs);
549 sigprocmask(SIG_BLOCK, &allsigs, &tmpsig);
557 log_print("no locking on control device");
560 error = device_user_lock(proc, &kbuf->i.lock);
563 case DLM_USER_UNLOCK:
565 log_print("no locking on control device");
568 error = device_user_unlock(proc, &kbuf->i.lock);
571 case DLM_USER_DEADLOCK:
573 log_print("no locking on control device");
576 error = device_user_deadlock(proc, &kbuf->i.lock);
579 case DLM_USER_CREATE_LOCKSPACE:
581 log_print("create/remove only on control device");
584 error = device_create_lockspace(&kbuf->i.lspace);
587 case DLM_USER_REMOVE_LOCKSPACE:
589 log_print("create/remove only on control device");
592 error = device_remove_lockspace(&kbuf->i.lspace);
597 log_print("no locking on control device");
600 error = device_user_purge(proc, &kbuf->i.purge);
604 log_print("Unknown command passed to DLM device : %d\n",
609 sigprocmask(SIG_SETMASK, &tmpsig, NULL);
616 /* Every process that opens the lockspace device has its own "proc" structure
617 hanging off the open file that's used to keep track of locks owned by the
618 process and asts that need to be delivered to the process. */
620 static int device_open(struct inode *inode, struct file *file)
622 struct dlm_user_proc *proc;
625 ls = dlm_find_lockspace_device(iminor(inode));
629 proc = kzalloc(sizeof(struct dlm_user_proc), GFP_KERNEL);
631 dlm_put_lockspace(ls);
635 proc->lockspace = ls->ls_local_handle;
636 INIT_LIST_HEAD(&proc->asts);
637 INIT_LIST_HEAD(&proc->locks);
638 INIT_LIST_HEAD(&proc->unlocking);
639 spin_lock_init(&proc->asts_spin);
640 spin_lock_init(&proc->locks_spin);
641 init_waitqueue_head(&proc->wait);
642 file->private_data = proc;
647 static int device_close(struct inode *inode, struct file *file)
649 struct dlm_user_proc *proc = file->private_data;
651 sigset_t tmpsig, allsigs;
653 ls = dlm_find_lockspace_local(proc->lockspace);
657 sigfillset(&allsigs);
658 sigprocmask(SIG_BLOCK, &allsigs, &tmpsig);
660 set_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags);
662 dlm_clear_proc_locks(ls, proc);
664 /* at this point no more lkb's should exist for this lockspace,
665 so there's no chance of dlm_user_add_ast() being called and
666 looking for lkb->ua->proc */
669 file->private_data = NULL;
671 dlm_put_lockspace(ls);
672 dlm_put_lockspace(ls); /* for the find in device_open() */
674 /* FIXME: AUTOFREE: if this ls is no longer used do
675 device_remove_lockspace() */
677 sigprocmask(SIG_SETMASK, &tmpsig, NULL);
683 static int copy_result_to_user(struct dlm_user_args *ua, int compat, int type,
684 int bmode, char __user *buf, size_t count)
687 struct dlm_lock_result32 result32;
689 struct dlm_lock_result result;
695 memset(&result, 0, sizeof(struct dlm_lock_result));
696 result.version[0] = DLM_DEVICE_VERSION_MAJOR;
697 result.version[1] = DLM_DEVICE_VERSION_MINOR;
698 result.version[2] = DLM_DEVICE_VERSION_PATCH;
699 memcpy(&result.lksb, &ua->lksb, sizeof(struct dlm_lksb));
700 result.user_lksb = ua->user_lksb;
702 /* FIXME: dlm1 provides for the user's bastparam/addr to not be updated
703 in a conversion unless the conversion is successful. See code
704 in dlm_user_convert() for updating ua from ua_tmp. OpenVMS, though,
705 notes that a new blocking AST address and parameter are set even if
706 the conversion fails, so maybe we should just do that. */
708 if (type == AST_BAST) {
709 result.user_astaddr = ua->bastaddr;
710 result.user_astparam = ua->bastparam;
711 result.bast_mode = bmode;
713 result.user_astaddr = ua->castaddr;
714 result.user_astparam = ua->castparam;
719 len = sizeof(struct dlm_lock_result32);
722 len = sizeof(struct dlm_lock_result);
725 /* copy lvb to userspace if there is one, it's been updated, and
726 the user buffer has space for it */
728 if (ua->update_user_lvb && ua->lksb.sb_lvbptr &&
729 count >= len + DLM_USER_LVB_LEN) {
730 if (copy_to_user(buf+len, ua->lksb.sb_lvbptr,
736 result.lvb_offset = len;
737 len += DLM_USER_LVB_LEN;
744 compat_output(&result, &result32);
745 resultptr = &result32;
749 if (copy_to_user(buf, resultptr, struct_len))
757 static int copy_version_to_user(char __user *buf, size_t count)
759 struct dlm_device_version ver;
761 memset(&ver, 0, sizeof(struct dlm_device_version));
762 ver.version[0] = DLM_DEVICE_VERSION_MAJOR;
763 ver.version[1] = DLM_DEVICE_VERSION_MINOR;
764 ver.version[2] = DLM_DEVICE_VERSION_PATCH;
766 if (copy_to_user(buf, &ver, sizeof(struct dlm_device_version)))
768 return sizeof(struct dlm_device_version);
771 /* a read returns a single ast described in a struct dlm_lock_result */
773 static ssize_t device_read(struct file *file, char __user *buf, size_t count,
776 struct dlm_user_proc *proc = file->private_data;
778 struct dlm_user_args *ua;
779 DECLARE_WAITQUEUE(wait, current);
780 int error, type=0, bmode=0, removed = 0;
782 if (count == sizeof(struct dlm_device_version)) {
783 error = copy_version_to_user(buf, count);
788 log_print("non-version read from control device %zu", count);
793 if (count < sizeof(struct dlm_lock_result32))
795 if (count < sizeof(struct dlm_lock_result))
799 /* do we really need this? can a read happen after a close? */
800 if (test_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags))
803 spin_lock(&proc->asts_spin);
804 if (list_empty(&proc->asts)) {
805 if (file->f_flags & O_NONBLOCK) {
806 spin_unlock(&proc->asts_spin);
810 add_wait_queue(&proc->wait, &wait);
813 set_current_state(TASK_INTERRUPTIBLE);
814 if (list_empty(&proc->asts) && !signal_pending(current)) {
815 spin_unlock(&proc->asts_spin);
817 spin_lock(&proc->asts_spin);
820 set_current_state(TASK_RUNNING);
821 remove_wait_queue(&proc->wait, &wait);
823 if (signal_pending(current)) {
824 spin_unlock(&proc->asts_spin);
829 /* there may be both completion and blocking asts to return for
830 the lkb, don't remove lkb from asts list unless no asts remain */
832 lkb = list_entry(proc->asts.next, struct dlm_lkb, lkb_astqueue);
834 if (lkb->lkb_ast_type & AST_COMP) {
835 lkb->lkb_ast_type &= ~AST_COMP;
837 } else if (lkb->lkb_ast_type & AST_BAST) {
838 lkb->lkb_ast_type &= ~AST_BAST;
840 bmode = lkb->lkb_bastmode;
843 if (!lkb->lkb_ast_type) {
844 list_del(&lkb->lkb_astqueue);
847 spin_unlock(&proc->asts_spin);
849 ua = (struct dlm_user_args *)lkb->lkb_astparam;
850 error = copy_result_to_user(ua,
851 test_bit(DLM_PROC_FLAGS_COMPAT, &proc->flags),
852 type, bmode, buf, count);
854 /* removes reference for the proc->asts lists added by
855 dlm_user_add_ast() and may result in the lkb being freed */
862 static unsigned int device_poll(struct file *file, poll_table *wait)
864 struct dlm_user_proc *proc = file->private_data;
866 poll_wait(file, &proc->wait, wait);
868 spin_lock(&proc->asts_spin);
869 if (!list_empty(&proc->asts)) {
870 spin_unlock(&proc->asts_spin);
871 return POLLIN | POLLRDNORM;
873 spin_unlock(&proc->asts_spin);
877 static int ctl_device_open(struct inode *inode, struct file *file)
879 file->private_data = NULL;
883 static int ctl_device_close(struct inode *inode, struct file *file)
888 static const struct file_operations device_fops = {
890 .release = device_close,
892 .write = device_write,
894 .owner = THIS_MODULE,
897 static const struct file_operations ctl_device_fops = {
898 .open = ctl_device_open,
899 .release = ctl_device_close,
901 .write = device_write,
902 .owner = THIS_MODULE,
905 int dlm_user_init(void)
909 ctl_device.name = "dlm-control";
910 ctl_device.fops = &ctl_device_fops;
911 ctl_device.minor = MISC_DYNAMIC_MINOR;
913 error = misc_register(&ctl_device);
915 log_print("misc_register failed for control device");
920 void dlm_user_exit(void)
922 misc_deregister(&ctl_device);